A review of semantic segmentation methods and their application in apple disease detection
Masoumeh Keshavarzi, Carl H. Mesarich, Donald G. Bailey, Martin Johnson, Gourab Sen Gupta
Abstract
Semantic segmentation, with pixel-wise classification, enables the precise identification of different parts of plants, as well as the diseases that occur on them, in agricultural images. Apples, as one of the most important fruit crops worldwide, are susceptible to various diseases, causing decreased crop quality and increased crop loss. To prevent disease progression and ensure prompt treatment, semantic segmentation acts as an effective method in the context of apple disease detection. This review provides a comprehensive analysis of semantic segmentation methods applied in apple disease detection, ranging from traditional approaches to state-of-the-art techniques. By systematically examining the entire pipeline, from dataset preparation to the segmentation and evaluation stages, this work not only synthesises existing knowledge but also reviews applied solutions and highlights remaining research gaps to enhance segmentation performance. Additionally, it offers a forward-looking perspective by proposing future research directions. Overall, this review aims to advance plant disease detection through semantic segmentation, with a particular emphasis on apples. • Semantic segmentation methods were reviewed for apple disease detection. • Techniques to enhance accuracy were reviewed in a detailed, step-by-step manner. • Tools and metrics for assessing plant disease severity were reviewed. • Costs of semantic segmentation were discussed. • Future research development was addressed through identifying gaps and solutions.